]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/hevc.c
lavu: Drop the {minus,plus}1 suffix from AVComponentDescriptor fields
[ffmpeg] / libavcodec / hevc.c
index f00667c6ff6d1135d0279d0c015cac4bef02ae77..a802e1e4ace4b44389342745d281e14600aaf667 100644 (file)
 
 #include "libavutil/attributes.h"
 #include "libavutil/common.h"
+#include "libavutil/display.h"
 #include "libavutil/internal.h"
 #include "libavutil/md5.h"
 #include "libavutil/opt.h"
 #include "libavutil/pixdesc.h"
+#include "libavutil/stereo3d.h"
 
+#include "bswapdsp.h"
 #include "bytestream.h"
 #include "cabac_functions.h"
-#include "dsputil.h"
 #include "golomb.h"
 #include "hevc.h"
 
@@ -80,20 +82,6 @@ static const uint8_t diag_scan2x2_inv[2][2] = {
     { 1, 3, },
 };
 
-const uint8_t ff_hevc_diag_scan4x4_x[16] = {
-    0, 0, 1, 0,
-    1, 2, 0, 1,
-    2, 3, 1, 2,
-    3, 2, 3, 3,
-};
-
-const uint8_t ff_hevc_diag_scan4x4_y[16] = {
-    0, 1, 0, 2,
-    1, 0, 3, 2,
-    1, 0, 3, 2,
-    1, 3, 2, 3,
-};
-
 static const uint8_t diag_scan4x4_inv[4][4] = {
     { 0,  2,  5,  9, },
     { 1,  4,  8, 12, },
@@ -101,44 +89,6 @@ static const uint8_t diag_scan4x4_inv[4][4] = {
     { 6, 10, 13, 15, },
 };
 
-const uint8_t ff_hevc_diag_scan8x8_x[64] = {
-    0, 0, 1, 0,
-    1, 2, 0, 1,
-    2, 3, 0, 1,
-    2, 3, 4, 0,
-    1, 2, 3, 4,
-    5, 0, 1, 2,
-    3, 4, 5, 6,
-    0, 1, 2, 3,
-    4, 5, 6, 7,
-    1, 2, 3, 4,
-    5, 6, 7, 2,
-    3, 4, 5, 6,
-    7, 3, 4, 5,
-    6, 7, 4, 5,
-    6, 7, 5, 6,
-    7, 6, 7, 7,
-};
-
-const uint8_t ff_hevc_diag_scan8x8_y[64] = {
-    0, 1, 0, 2,
-    1, 0, 3, 2,
-    1, 0, 4, 3,
-    2, 1, 0, 5,
-    4, 3, 2, 1,
-    0, 6, 5, 4,
-    3, 2, 1, 0,
-    7, 6, 5, 4,
-    3, 2, 1, 0,
-    7, 6, 5, 4,
-    3, 2, 1, 7,
-    6, 5, 4, 3,
-    2, 7, 6, 5,
-    4, 3, 7, 6,
-    5, 4, 7, 6,
-    5, 7, 6, 7,
-};
-
 static const uint8_t diag_scan8x8_inv[8][8] = {
     {  0,  2,  5,  9, 14, 20, 27, 35, },
     {  1,  4,  8, 13, 19, 26, 34, 42, },
@@ -164,7 +114,6 @@ static void pic_arrays_free(HEVCContext *s)
 {
     av_freep(&s->sao);
     av_freep(&s->deblock);
-    av_freep(&s->split_cu_flag);
 
     av_freep(&s->skip_flag);
     av_freep(&s->tab_ct_depth);
@@ -190,7 +139,6 @@ static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
     int log2_min_cb_size = sps->log2_min_cb_size;
     int width            = sps->width;
     int height           = sps->height;
-    int pic_size         = width * height;
     int pic_size_in_ctb  = ((width  >> log2_min_cb_size) + 1) *
                            ((height >> log2_min_cb_size) + 1);
     int ctb_count        = sps->ctb_width * sps->ctb_height;
@@ -201,8 +149,7 @@ static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
 
     s->sao           = av_mallocz_array(ctb_count, sizeof(*s->sao));
     s->deblock       = av_mallocz_array(ctb_count, sizeof(*s->deblock));
-    s->split_cu_flag = av_malloc(pic_size);
-    if (!s->sao || !s->deblock || !s->split_cu_flag)
+    if (!s->sao || !s->deblock)
         goto fail;
 
     s->skip_flag    = av_malloc(pic_size_in_ctb);
@@ -211,7 +158,7 @@ static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
         goto fail;
 
     s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
-    s->tab_ipm  = av_malloc(min_pu_size);
+    s->tab_ipm  = av_mallocz(min_pu_size);
     s->is_pcm   = av_malloc(min_pu_size);
     if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
         goto fail;
@@ -252,10 +199,10 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
     uint8_t luma_weight_l1_flag[16];
     uint8_t chroma_weight_l1_flag[16];
 
-    s->sh.luma_log2_weight_denom = get_ue_golomb_long(gb);
-    if (s->sps->chroma_format_idc != 0) {
+    s->sh.luma_log2_weight_denom = av_clip(get_ue_golomb_long(gb), 0, 7);
+    if (s->ps.sps->chroma_format_idc != 0) {
         int delta = get_se_golomb(gb);
-        s->sh.chroma_log2_weight_denom = av_clip_c(s->sh.luma_log2_weight_denom + delta, 0, 7);
+        s->sh.chroma_log2_weight_denom = av_clip(s->sh.luma_log2_weight_denom + delta, 0, 7);
     }
 
     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
@@ -265,7 +212,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
             s->sh.luma_offset_l0[i] = 0;
         }
     }
-    if (s->sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
+    if (s->ps.sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
         for (i = 0; i < s->sh.nb_refs[L0]; i++)
             chroma_weight_l0_flag[i] = get_bits1(gb);
     } else {
@@ -283,7 +230,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
                 int delta_chroma_weight_l0 = get_se_golomb(gb);
                 int delta_chroma_offset_l0 = get_se_golomb(gb);
                 s->sh.chroma_weight_l0[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l0;
-                s->sh.chroma_offset_l0[i][j] = av_clip_c((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
+                s->sh.chroma_offset_l0[i][j] = av_clip((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
                                                                                     >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
             }
         } else {
@@ -301,7 +248,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
                 s->sh.luma_offset_l1[i] = 0;
             }
         }
-        if (s->sps->chroma_format_idc != 0) {
+        if (s->ps.sps->chroma_format_idc != 0) {
             for (i = 0; i < s->sh.nb_refs[L1]; i++)
                 chroma_weight_l1_flag[i] = get_bits1(gb);
         } else {
@@ -319,7 +266,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
                     int delta_chroma_weight_l1 = get_se_golomb(gb);
                     int delta_chroma_offset_l1 = get_se_golomb(gb);
                     s->sh.chroma_weight_l1[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l1;
-                    s->sh.chroma_offset_l1[i][j] = av_clip_c((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
+                    s->sh.chroma_offset_l1[i][j] = av_clip((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
                                                                                         >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
                 }
             } else {
@@ -334,10 +281,10 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
 
 static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
 {
-    const HEVCSPS *sps = s->sps;
+    const HEVCSPS *sps = s->ps.sps;
     int max_poc_lsb    = 1 << sps->log2_max_poc_lsb;
     int prev_delta_msb = 0;
-    int nb_sps = 0, nb_sh;
+    unsigned int nb_sps = 0, nb_sh;
     int i;
 
     rps->nb_refs = 0;
@@ -384,44 +331,93 @@ static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
     return 0;
 }
 
+static void export_stream_params(AVCodecContext *avctx, const HEVCParamSets *ps,
+                                 const HEVCSPS *sps)
+{
+    const HEVCVPS *vps = (const HEVCVPS*)ps->vps_list[sps->vps_id]->data;
+    unsigned int num = 0, den = 0;
+
+    avctx->pix_fmt             = sps->pix_fmt;
+    avctx->coded_width         = sps->width;
+    avctx->coded_height        = sps->height;
+    avctx->width               = sps->output_width;
+    avctx->height              = sps->output_height;
+    avctx->has_b_frames        = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
+    avctx->profile             = sps->ptl.general_ptl.profile_idc;
+    avctx->level               = sps->ptl.general_ptl.level_idc;
+
+    ff_set_sar(avctx, sps->vui.sar);
+
+    if (sps->vui.video_signal_type_present_flag)
+        avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
+                                                            : AVCOL_RANGE_MPEG;
+    else
+        avctx->color_range = AVCOL_RANGE_MPEG;
+
+    if (sps->vui.colour_description_present_flag) {
+        avctx->color_primaries = sps->vui.colour_primaries;
+        avctx->color_trc       = sps->vui.transfer_characteristic;
+        avctx->colorspace      = sps->vui.matrix_coeffs;
+    } else {
+        avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
+        avctx->color_trc       = AVCOL_TRC_UNSPECIFIED;
+        avctx->colorspace      = AVCOL_SPC_UNSPECIFIED;
+    }
+
+    if (vps->vps_timing_info_present_flag) {
+        num = vps->vps_num_units_in_tick;
+        den = vps->vps_time_scale;
+    } else if (sps->vui.vui_timing_info_present_flag) {
+        num = sps->vui.vui_num_units_in_tick;
+        den = sps->vui.vui_time_scale;
+    }
+
+    if (num != 0 && den != 0)
+        av_reduce(&avctx->framerate.den, &avctx->framerate.num,
+                  num, den, 1 << 30);
+}
+
 static int set_sps(HEVCContext *s, const HEVCSPS *sps)
 {
+    #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL)
+    enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
     int ret;
 
     pic_arrays_free(s);
+    s->ps.sps = NULL;
+    s->ps.vps = NULL;
+
+    if (!sps)
+        return 0;
+
     ret = pic_arrays_init(s, sps);
     if (ret < 0)
         goto fail;
 
-    s->avctx->coded_width         = sps->width;
-    s->avctx->coded_height        = sps->height;
-    s->avctx->width               = sps->output_width;
-    s->avctx->height              = sps->output_height;
-    s->avctx->pix_fmt             = sps->pix_fmt;
-    s->avctx->sample_aspect_ratio = sps->vui.sar;
-    s->avctx->has_b_frames        = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
-
-    if (sps->vui.video_signal_type_present_flag)
-        s->avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
-                                                               : AVCOL_RANGE_MPEG;
-    else
-        s->avctx->color_range = AVCOL_RANGE_MPEG;
+    export_stream_params(s->avctx, &s->ps, sps);
 
-    if (sps->vui.colour_description_present_flag) {
-        s->avctx->color_primaries = sps->vui.colour_primaries;
-        s->avctx->color_trc       = sps->vui.transfer_characteristic;
-        s->avctx->colorspace      = sps->vui.matrix_coeffs;
-    } else {
-        s->avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
-        s->avctx->color_trc       = AVCOL_TRC_UNSPECIFIED;
-        s->avctx->colorspace      = AVCOL_SPC_UNSPECIFIED;
+    if (sps->pix_fmt == AV_PIX_FMT_YUV420P || sps->pix_fmt == AV_PIX_FMT_YUVJ420P) {
+#if CONFIG_HEVC_DXVA2_HWACCEL
+        *fmt++ = AV_PIX_FMT_DXVA2_VLD;
+#endif
+#if CONFIG_HEVC_D3D11VA_HWACCEL
+        *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
+#endif
     }
 
+    *fmt++ = sps->pix_fmt;
+    *fmt = AV_PIX_FMT_NONE;
+
+    ret = ff_get_format(s->avctx, pix_fmts);
+    if (ret < 0)
+        goto fail;
+    s->avctx->pix_fmt = ret;
+
     ff_hevc_pred_init(&s->hpc,     sps->bit_depth);
     ff_hevc_dsp_init (&s->hevcdsp, sps->bit_depth);
     ff_videodsp_init (&s->vdsp,    sps->bit_depth);
 
-    if (sps->sao_enabled) {
+    if (sps->sao_enabled && !s->avctx->hwaccel) {
         av_frame_unref(s->tmp_frame);
         ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
         if (ret < 0)
@@ -429,13 +425,14 @@ static int set_sps(HEVCContext *s, const HEVCSPS *sps)
         s->frame = s->tmp_frame;
     }
 
-    s->sps = sps;
-    s->vps = s->vps_list[s->sps->vps_id];
+    s->ps.sps = sps;
+    s->ps.vps = (HEVCVPS*) s->ps.vps_list[s->ps.sps->vps_id]->data;
+
     return 0;
 
 fail:
     pic_arrays_free(s);
-    s->sps = NULL;
+    s->ps.sps = NULL;
     return ret;
 }
 
@@ -453,26 +450,26 @@ static int hls_slice_header(HEVCContext *s)
         if (IS_IDR(s))
             ff_hevc_clear_refs(s);
     }
-    if (s->nal_unit_type >= 16 && s->nal_unit_type <= 23)
+    if (IS_IRAP(s))
         sh->no_output_of_prior_pics_flag = get_bits1(gb);
 
     sh->pps_id = get_ue_golomb_long(gb);
-    if (sh->pps_id >= MAX_PPS_COUNT || !s->pps_list[sh->pps_id]) {
+    if (sh->pps_id >= MAX_PPS_COUNT || !s->ps.pps_list[sh->pps_id]) {
         av_log(s->avctx, AV_LOG_ERROR, "PPS id out of range: %d\n", sh->pps_id);
         return AVERROR_INVALIDDATA;
     }
     if (!sh->first_slice_in_pic_flag &&
-        s->pps != (HEVCPPS*)s->pps_list[sh->pps_id]->data) {
+        s->ps.pps != (HEVCPPS*)s->ps.pps_list[sh->pps_id]->data) {
         av_log(s->avctx, AV_LOG_ERROR, "PPS changed between slices.\n");
         return AVERROR_INVALIDDATA;
     }
-    s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
+    s->ps.pps = (HEVCPPS*)s->ps.pps_list[sh->pps_id]->data;
 
-    if (s->sps != (HEVCSPS*)s->sps_list[s->pps->sps_id]->data) {
-        s->sps = (HEVCSPS*)s->sps_list[s->pps->sps_id]->data;
+    if (s->ps.sps != (HEVCSPS*)s->ps.sps_list[s->ps.pps->sps_id]->data) {
+        s->ps.sps = (HEVCSPS*)s->ps.sps_list[s->ps.pps->sps_id]->data;
 
         ff_hevc_clear_refs(s);
-        ret = set_sps(s, s->sps);
+        ret = set_sps(s, s->ps.sps);
         if (ret < 0)
             return ret;
 
@@ -484,13 +481,13 @@ static int hls_slice_header(HEVCContext *s)
     if (!sh->first_slice_in_pic_flag) {
         int slice_address_length;
 
-        if (s->pps->dependent_slice_segments_enabled_flag)
+        if (s->ps.pps->dependent_slice_segments_enabled_flag)
             sh->dependent_slice_segment_flag = get_bits1(gb);
 
-        slice_address_length = av_ceil_log2(s->sps->ctb_width *
-                                            s->sps->ctb_height);
-        sh->slice_segment_addr = get_bits(gb, slice_address_length);
-        if (sh->slice_segment_addr >= s->sps->ctb_width * s->sps->ctb_height) {
+        slice_address_length = av_ceil_log2(s->ps.sps->ctb_width *
+                                            s->ps.sps->ctb_height);
+        sh->slice_segment_addr = slice_address_length ? get_bits(gb, slice_address_length) : 0;
+        if (sh->slice_segment_addr >= s->ps.sps->ctb_width * s->ps.sps->ctb_height) {
             av_log(s->avctx, AV_LOG_ERROR,
                    "Invalid slice segment address: %u.\n",
                    sh->slice_segment_addr);
@@ -510,7 +507,7 @@ static int hls_slice_header(HEVCContext *s)
     if (!sh->dependent_slice_segment_flag) {
         s->slice_initialized = 0;
 
-        for (i = 0; i < s->pps->num_extra_slice_header_bits; i++)
+        for (i = 0; i < s->ps.pps->num_extra_slice_header_bits; i++)
             skip_bits(gb, 1);  // slice_reserved_undetermined_flag[]
 
         sh->slice_type = get_ue_golomb_long(gb);
@@ -526,16 +523,18 @@ static int hls_slice_header(HEVCContext *s)
             return AVERROR_INVALIDDATA;
         }
 
-        if (s->pps->output_flag_present_flag)
+        // when flag is not present, picture is inferred to be output
+        sh->pic_output_flag = 1;
+        if (s->ps.pps->output_flag_present_flag)
             sh->pic_output_flag = get_bits1(gb);
 
-        if (s->sps->separate_colour_plane_flag)
+        if (s->ps.sps->separate_colour_plane_flag)
             sh->colour_plane_id = get_bits(gb, 2);
 
         if (!IS_IDR(s)) {
-            int short_term_ref_pic_set_sps_flag, poc;
+            int poc;
 
-            sh->pic_order_cnt_lsb = get_bits(gb, s->sps->log2_max_poc_lsb);
+            sh->pic_order_cnt_lsb = get_bits(gb, s->ps.sps->log2_max_poc_lsb);
             poc = ff_hevc_compute_poc(s, sh->pic_order_cnt_lsb);
             if (!sh->first_slice_in_pic_flag && poc != s->poc) {
                 av_log(s->avctx, AV_LOG_WARNING,
@@ -546,24 +545,26 @@ static int hls_slice_header(HEVCContext *s)
             }
             s->poc = poc;
 
-            short_term_ref_pic_set_sps_flag = get_bits1(gb);
-            if (!short_term_ref_pic_set_sps_flag) {
-                ret = ff_hevc_decode_short_term_rps(s, &sh->slice_rps, s->sps, 1);
+            sh->short_term_ref_pic_set_sps_flag = get_bits1(gb);
+            if (!sh->short_term_ref_pic_set_sps_flag) {
+                int pos = get_bits_left(gb);
+                ret = ff_hevc_decode_short_term_rps(gb, s->avctx, &sh->slice_rps, s->ps.sps, 1);
                 if (ret < 0)
                     return ret;
 
+                sh->short_term_ref_pic_set_size = pos - get_bits_left(gb);
                 sh->short_term_rps = &sh->slice_rps;
             } else {
                 int numbits, rps_idx;
 
-                if (!s->sps->nb_st_rps) {
+                if (!s->ps.sps->nb_st_rps) {
                     av_log(s->avctx, AV_LOG_ERROR, "No ref lists in the SPS.\n");
                     return AVERROR_INVALIDDATA;
                 }
 
-                numbits = av_ceil_log2(s->sps->nb_st_rps);
+                numbits = av_ceil_log2(s->ps.sps->nb_st_rps);
                 rps_idx = numbits > 0 ? get_bits(gb, numbits) : 0;
-                sh->short_term_rps = &s->sps->st_rps[rps_idx];
+                sh->short_term_rps = &s->ps.sps->st_rps[rps_idx];
             }
 
             ret = decode_lt_rps(s, &sh->long_term_rps, gb);
@@ -573,7 +574,7 @@ static int hls_slice_header(HEVCContext *s)
                     return AVERROR_INVALIDDATA;
             }
 
-            if (s->sps->sps_temporal_mvp_enabled_flag)
+            if (s->ps.sps->sps_temporal_mvp_enabled_flag)
                 sh->slice_temporal_mvp_enabled_flag = get_bits1(gb);
             else
                 sh->slice_temporal_mvp_enabled_flag = 0;
@@ -593,7 +594,7 @@ static int hls_slice_header(HEVCContext *s)
             s->nal_unit_type != NAL_RASL_R)
             s->pocTid0 = s->poc;
 
-        if (s->sps->sao_enabled) {
+        if (s->ps.sps->sao_enabled) {
             sh->slice_sample_adaptive_offset_flag[0] = get_bits1(gb);
             sh->slice_sample_adaptive_offset_flag[1] =
             sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
@@ -607,9 +608,9 @@ static int hls_slice_header(HEVCContext *s)
         if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
             int nb_refs;
 
-            sh->nb_refs[L0] = s->pps->num_ref_idx_l0_default_active;
+            sh->nb_refs[L0] = s->ps.pps->num_ref_idx_l0_default_active;
             if (sh->slice_type == B_SLICE)
-                sh->nb_refs[L1] = s->pps->num_ref_idx_l1_default_active;
+                sh->nb_refs[L1] = s->ps.pps->num_ref_idx_l1_default_active;
 
             if (get_bits1(gb)) { // num_ref_idx_active_override_flag
                 sh->nb_refs[L0] = get_ue_golomb_long(gb) + 1;
@@ -630,7 +631,7 @@ static int hls_slice_header(HEVCContext *s)
                 return AVERROR_INVALIDDATA;
             }
 
-            if (s->pps->lists_modification_present_flag && nb_refs > 1) {
+            if (s->ps.pps->lists_modification_present_flag && nb_refs > 1) {
                 sh->rpl_modification_flag[0] = get_bits1(gb);
                 if (sh->rpl_modification_flag[0]) {
                     for (i = 0; i < sh->nb_refs[L0]; i++)
@@ -648,7 +649,7 @@ static int hls_slice_header(HEVCContext *s)
             if (sh->slice_type == B_SLICE)
                 sh->mvd_l1_zero_flag = get_bits1(gb);
 
-            if (s->pps->cabac_init_present_flag)
+            if (s->ps.pps->cabac_init_present_flag)
                 sh->cabac_init_flag = get_bits1(gb);
             else
                 sh->cabac_init_flag = 0;
@@ -670,8 +671,8 @@ static int hls_slice_header(HEVCContext *s)
                 }
             }
 
-            if ((s->pps->weighted_pred_flag   && sh->slice_type == P_SLICE) ||
-                (s->pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
+            if ((s->ps.pps->weighted_pred_flag   && sh->slice_type == P_SLICE) ||
+                (s->ps.pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
                 pred_weight_table(s, gb);
             }
 
@@ -685,7 +686,8 @@ static int hls_slice_header(HEVCContext *s)
         }
 
         sh->slice_qp_delta = get_se_golomb(gb);
-        if (s->pps->pic_slice_level_chroma_qp_offsets_present_flag) {
+
+        if (s->ps.pps->pic_slice_level_chroma_qp_offsets_present_flag) {
             sh->slice_cb_qp_offset = get_se_golomb(gb);
             sh->slice_cr_qp_offset = get_se_golomb(gb);
         } else {
@@ -693,10 +695,10 @@ static int hls_slice_header(HEVCContext *s)
             sh->slice_cr_qp_offset = 0;
         }
 
-        if (s->pps->deblocking_filter_control_present_flag) {
+        if (s->ps.pps->deblocking_filter_control_present_flag) {
             int deblocking_filter_override_flag = 0;
 
-            if (s->pps->deblocking_filter_override_enabled_flag)
+            if (s->ps.pps->deblocking_filter_override_enabled_flag)
                 deblocking_filter_override_flag = get_bits1(gb);
 
             if (deblocking_filter_override_flag) {
@@ -706,9 +708,9 @@ static int hls_slice_header(HEVCContext *s)
                     sh->tc_offset   = get_se_golomb(gb) * 2;
                 }
             } else {
-                sh->disable_deblocking_filter_flag = s->pps->disable_dbf;
-                sh->beta_offset                    = s->pps->beta_offset;
-                sh->tc_offset                      = s->pps->tc_offset;
+                sh->disable_deblocking_filter_flag = s->ps.pps->disable_dbf;
+                sh->beta_offset                    = s->ps.pps->beta_offset;
+                sh->tc_offset                      = s->ps.pps->tc_offset;
             }
         } else {
             sh->disable_deblocking_filter_flag = 0;
@@ -716,13 +718,13 @@ static int hls_slice_header(HEVCContext *s)
             sh->tc_offset                      = 0;
         }
 
-        if (s->pps->seq_loop_filter_across_slices_enabled_flag &&
+        if (s->ps.pps->seq_loop_filter_across_slices_enabled_flag &&
             (sh->slice_sample_adaptive_offset_flag[0] ||
              sh->slice_sample_adaptive_offset_flag[1] ||
              !sh->disable_deblocking_filter_flag)) {
             sh->slice_loop_filter_across_slices_enabled_flag = get_bits1(gb);
         } else {
-            sh->slice_loop_filter_across_slices_enabled_flag = s->pps->seq_loop_filter_across_slices_enabled_flag;
+            sh->slice_loop_filter_across_slices_enabled_flag = s->ps.pps->seq_loop_filter_across_slices_enabled_flag;
         }
     } else if (!s->slice_initialized) {
         av_log(s->avctx, AV_LOG_ERROR, "Independent slice segment missing.\n");
@@ -730,7 +732,7 @@ static int hls_slice_header(HEVCContext *s)
     }
 
     sh->num_entry_point_offsets = 0;
-    if (s->pps->tiles_enabled_flag || s->pps->entropy_coding_sync_enabled_flag) {
+    if (s->ps.pps->tiles_enabled_flag || s->ps.pps->entropy_coding_sync_enabled_flag) {
         sh->num_entry_point_offsets = get_ue_golomb_long(gb);
         if (sh->num_entry_point_offsets > 0) {
             int offset_len = get_ue_golomb_long(gb) + 1;
@@ -740,28 +742,43 @@ static int hls_slice_header(HEVCContext *s)
         }
     }
 
-    if (s->pps->slice_header_extension_present_flag) {
-        int length = get_ue_golomb_long(gb);
+    if (s->ps.pps->slice_header_extension_present_flag) {
+        unsigned int length = get_ue_golomb_long(gb);
         for (i = 0; i < length; i++)
             skip_bits(gb, 8);  // slice_header_extension_data_byte
     }
 
     // Inferred parameters
-    sh->slice_qp          = 26 + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
+    sh->slice_qp = 26 + s->ps.pps->pic_init_qp_minus26 + sh->slice_qp_delta;
+    if (sh->slice_qp > 51 ||
+        sh->slice_qp < -s->ps.sps->qp_bd_offset) {
+        av_log(s->avctx, AV_LOG_ERROR,
+               "The slice_qp %d is outside the valid range "
+               "[%d, 51].\n",
+               sh->slice_qp,
+               -s->ps.sps->qp_bd_offset);
+        return AVERROR_INVALIDDATA;
+    }
+
     sh->slice_ctb_addr_rs = sh->slice_segment_addr;
 
+    if (!s->sh.slice_ctb_addr_rs && s->sh.dependent_slice_segment_flag) {
+        av_log(s->avctx, AV_LOG_ERROR, "Impossible slice segment.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
     s->HEVClc.first_qp_group = !s->sh.dependent_slice_segment_flag;
 
-    if (!s->pps->cu_qp_delta_enabled_flag)
-        s->HEVClc.qp_y = ((s->sh.slice_qp + 52 + 2 * s->sps->qp_bd_offset) %
-                          (52 + s->sps->qp_bd_offset)) - s->sps->qp_bd_offset;
+    if (!s->ps.pps->cu_qp_delta_enabled_flag)
+        s->HEVClc.qp_y = FFUMOD(s->sh.slice_qp + 52 + 2 * s->ps.sps->qp_bd_offset,
+                                52 + s->ps.sps->qp_bd_offset) - s->ps.sps->qp_bd_offset;
 
     s->slice_initialized = 1;
 
     return 0;
 }
 
-#define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
+#define CTB(tab, x, y) ((tab)[(y) * s->ps.sps->ctb_width + (x)])
 
 #define SET_SAO(elem, value)                            \
 do {                                                    \
@@ -780,7 +797,7 @@ static void hls_sao_param(HEVCContext *s, int rx, int ry)
     HEVCLocalContext *lc    = &s->HEVClc;
     int sao_merge_left_flag = 0;
     int sao_merge_up_flag   = 0;
-    int shift               = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
+    int shift               = s->ps.sps->bit_depth - FFMIN(s->ps.sps->bit_depth, 10);
     SAOParams *sao          = &CTB(s->sao, rx, ry);
     int c_idx, i;
 
@@ -870,10 +887,10 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
     const uint8_t *scan_x_cg, *scan_y_cg, *scan_x_off, *scan_y_off;
 
     ptrdiff_t stride = s->frame->linesize[c_idx];
-    int hshift       = s->sps->hshift[c_idx];
-    int vshift       = s->sps->vshift[c_idx];
+    int hshift       = s->ps.sps->hshift[c_idx];
+    int vshift       = s->ps.sps->vshift[c_idx];
     uint8_t *dst     = &s->frame->data[c_idx][(y0 >> vshift) * stride +
-                                              ((x0 >> hshift) << s->sps->pixel_shift)];
+                                              ((x0 >> hshift) << s->ps.sps->pixel_shift)];
     DECLARE_ALIGNED(16, int16_t, coeffs[MAX_TB_SIZE * MAX_TB_SIZE]) = { 0 };
     DECLARE_ALIGNED(8, uint8_t, significant_coeff_group_flag[8][8]) = { { 0 } };
 
@@ -903,16 +920,16 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
         int qp_y = lc->qp_y;
 
         if (c_idx == 0) {
-            qp = qp_y + s->sps->qp_bd_offset;
+            qp = qp_y + s->ps.sps->qp_bd_offset;
         } else {
             int qp_i, offset;
 
             if (c_idx == 1)
-                offset = s->pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
+                offset = s->ps.pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
             else
-                offset = s->pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
+                offset = s->ps.pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
 
-            qp_i = av_clip_c(qp_y + offset, -s->sps->qp_bd_offset, 57);
+            qp_i = av_clip(qp_y + offset, -s->ps.sps->qp_bd_offset, 57);
             if (qp_i < 30)
                 qp = qp_i;
             else if (qp_i > 43)
@@ -920,18 +937,18 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
             else
                 qp = qp_c[qp_i - 30];
 
-            qp += s->sps->qp_bd_offset;
+            qp += s->ps.sps->qp_bd_offset;
         }
 
-        shift    = s->sps->bit_depth + log2_trafo_size - 5;
+        shift    = s->ps.sps->bit_depth + log2_trafo_size - 5;
         add      = 1 << (shift - 1);
         scale    = level_scale[rem6[qp]] << (div6[qp]);
         scale_m  = 16; // default when no custom scaling lists.
         dc_scale = 16;
 
-        if (s->sps->scaling_list_enable_flag) {
-            const ScalingList *sl = s->pps->scaling_list_data_present_flag ?
-                                    &s->pps->scaling_list : &s->sps->scaling_list;
+        if (s->ps.sps->scaling_list_enable_flag) {
+            const ScalingList *sl = s->ps.pps->scaling_list_data_present_flag ?
+                                    &s->ps.pps->scaling_list : &s->ps.sps->scaling_list;
             int matrix_id = lc->cu.pred_mode != MODE_INTRA;
 
             if (log2_trafo_size != 5)
@@ -943,7 +960,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
         }
     }
 
-    if (s->pps->transform_skip_enabled_flag &&
+    if (s->ps.pps->transform_skip_enabled_flag &&
         !lc->cu.cu_transquant_bypass_flag   &&
         log2_trafo_size == 2) {
         transform_skip_flag = ff_hevc_transform_skip_flag_decode(s, c_idx);
@@ -1127,7 +1144,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
             if (first_greater1_coeff_idx != -1) {
                 coeff_abs_level_greater1_flag[first_greater1_coeff_idx] += ff_hevc_coeff_abs_level_greater2_flag_decode(s, c_idx, ctx_set);
             }
-            if (!s->pps->sign_data_hiding_flag || !sign_hidden) {
+            if (!s->ps.pps->sign_data_hiding_flag || !sign_hidden) {
                 coeff_sign_flag = ff_hevc_coeff_sign_flag(s, nb_significant_coeff_flag) << (16 - nb_significant_coeff_flag);
             } else {
                 coeff_sign_flag = ff_hevc_coeff_sign_flag(s, nb_significant_coeff_flag - 1) << (16 - (nb_significant_coeff_flag - 1));
@@ -1145,7 +1162,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
                     if ((trans_coeff_level) > (3 * (1 << c_rice_param)))
                         c_rice_param = FFMIN(c_rice_param + 1, 4);
                 }
-                if (s->pps->sign_data_hiding_flag && sign_hidden) {
+                if (s->ps.pps->sign_data_hiding_flag && sign_hidden) {
                     sum_abs += trans_coeff_level;
                     if (n == first_nz_pos_in_cg && ((sum_abs & 1) == 1))
                         trans_coeff_level = -trans_coeff_level;
@@ -1154,7 +1171,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
                     trans_coeff_level = -trans_coeff_level;
                 coeff_sign_flag <<= 1;
                 if (!lc->cu.cu_transquant_bypass_flag) {
-                    if (s->sps->scaling_list_enable_flag) {
+                    if (s->ps.sps->scaling_list_enable_flag) {
                         if (y_c || x_c || log2_trafo_size < 4) {
                             int pos;
                             switch (log2_trafo_size) {
@@ -1195,10 +1212,10 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
     }
 }
 
-static void hls_transform_unit(HEVCContext *s, int x0, int y0,
-                               int xBase, int yBase, int cb_xBase, int cb_yBase,
-                               int log2_cb_size, int log2_trafo_size,
-                               int trafo_depth, int blk_idx)
+static int hls_transform_unit(HEVCContext *s, int x0, int y0,
+                              int xBase, int yBase, int cb_xBase, int cb_yBase,
+                              int log2_cb_size, int log2_trafo_size,
+                              int blk_idx, int cbf_luma, int cbf_cb, int cbf_cr)
 {
     HEVCLocalContext *lc = &s->HEVClc;
 
@@ -1206,33 +1223,43 @@ static void hls_transform_unit(HEVCContext *s, int x0, int y0,
         int trafo_size = 1 << log2_trafo_size;
         ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
 
-        s->hpc.intra_pred(s, x0, y0, log2_trafo_size, 0);
+        s->hpc.intra_pred[log2_trafo_size - 2](s, x0, y0, 0);
         if (log2_trafo_size > 2) {
-            trafo_size = trafo_size << (s->sps->hshift[1] - 1);
+            trafo_size = trafo_size << (s->ps.sps->hshift[1] - 1);
             ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
-            s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 1);
-            s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 2);
+            s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 1);
+            s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 2);
         } else if (blk_idx == 3) {
-            trafo_size = trafo_size << s->sps->hshift[1];
+            trafo_size = trafo_size << s->ps.sps->hshift[1];
             ff_hevc_set_neighbour_available(s, xBase, yBase,
                                             trafo_size, trafo_size);
-            s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 1);
-            s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 2);
+            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
+            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
         }
     }
 
-    if (lc->tt.cbf_luma ||
-        SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
-        SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
+    if (cbf_luma || cbf_cb || cbf_cr) {
         int scan_idx   = SCAN_DIAG;
         int scan_idx_c = SCAN_DIAG;
 
-        if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
+        if (s->ps.pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
             lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
             if (lc->tu.cu_qp_delta != 0)
                 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
                     lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
             lc->tu.is_cu_qp_delta_coded = 1;
+
+            if (lc->tu.cu_qp_delta < -(26 + s->ps.sps->qp_bd_offset / 2) ||
+                lc->tu.cu_qp_delta >  (25 + s->ps.sps->qp_bd_offset / 2)) {
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "The cu_qp_delta %d is outside the valid range "
+                       "[%d, %d].\n",
+                       lc->tu.cu_qp_delta,
+                       -(26 + s->ps.sps->qp_bd_offset / 2),
+                        (25 + s->ps.sps->qp_bd_offset / 2));
+                return AVERROR_INVALIDDATA;
+            }
+
             ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
         }
 
@@ -1254,30 +1281,31 @@ static void hls_transform_unit(HEVCContext *s, int x0, int y0,
             }
         }
 
-        if (lc->tt.cbf_luma)
+        if (cbf_luma)
             hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
         if (log2_trafo_size > 2) {
-            if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0))
+            if (cbf_cb)
                 hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
-            if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0))
+            if (cbf_cr)
                 hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
         } else if (blk_idx == 3) {
-            if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], xBase, yBase))
+            if (cbf_cb)
                 hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
-            if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], xBase, yBase))
+            if (cbf_cr)
                 hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
         }
     }
+    return 0;
 }
 
 static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
 {
     int cb_size          = 1 << log2_cb_size;
-    int log2_min_pu_size = s->sps->log2_min_pu_size;
+    int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
 
-    int min_pu_width     = s->sps->min_pu_width;
-    int x_end = FFMIN(x0 + cb_size, s->sps->width);
-    int y_end = FFMIN(y0 + cb_size, s->sps->height);
+    int min_pu_width     = s->ps.sps->min_pu_width;
+    int x_end = FFMIN(x0 + cb_size, s->ps.sps->width);
+    int y_end = FFMIN(y0 + cb_size, s->ps.sps->height);
     int i, j;
 
     for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
@@ -1285,23 +1313,15 @@ static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_si
             s->is_pcm[i + j * min_pu_width] = 2;
 }
 
-static void hls_transform_tree(HEVCContext *s, int x0, int y0,
-                               int xBase, int yBase, int cb_xBase, int cb_yBase,
-                               int log2_cb_size, int log2_trafo_size,
-                               int trafo_depth, int blk_idx)
+static int hls_transform_tree(HEVCContext *s, int x0, int y0,
+                              int xBase, int yBase, int cb_xBase, int cb_yBase,
+                              int log2_cb_size, int log2_trafo_size,
+                              int trafo_depth, int blk_idx,
+                              int cbf_cb, int cbf_cr)
 {
     HEVCLocalContext *lc = &s->HEVClc;
     uint8_t split_transform_flag;
-
-    if (trafo_depth > 0 && log2_trafo_size == 2) {
-        SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
-            SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase);
-        SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
-            SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase);
-    } else {
-        SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
-        SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) = 0;
-    }
+    int ret;
 
     if (lc->cu.intra_split_flag) {
         if (trafo_depth == 1)
@@ -1310,66 +1330,68 @@ static void hls_transform_tree(HEVCContext *s, int x0, int y0,
         lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
     }
 
-    lc->tt.cbf_luma = 1;
-
-    lc->tt.inter_split_flag = s->sps->max_transform_hierarchy_depth_inter == 0 &&
-                              lc->cu.pred_mode == MODE_INTER &&
-                              lc->cu.part_mode != PART_2Nx2N &&
-                              trafo_depth == 0;
-
-    if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
-        log2_trafo_size >  s->sps->log2_min_tb_size    &&
+    if (log2_trafo_size <= s->ps.sps->log2_max_trafo_size &&
+        log2_trafo_size >  s->ps.sps->log2_min_tb_size    &&
         trafo_depth     < lc->cu.max_trafo_depth       &&
         !(lc->cu.intra_split_flag && trafo_depth == 0)) {
         split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
     } else {
-        split_transform_flag = log2_trafo_size > s->sps->log2_max_trafo_size ||
+        int inter_split = s->ps.sps->max_transform_hierarchy_depth_inter == 0 &&
+                          lc->cu.pred_mode == MODE_INTER &&
+                          lc->cu.part_mode != PART_2Nx2N &&
+                          trafo_depth == 0;
+
+        split_transform_flag = log2_trafo_size > s->ps.sps->log2_max_trafo_size ||
                                (lc->cu.intra_split_flag && trafo_depth == 0) ||
-                               lc->tt.inter_split_flag;
+                               inter_split;
     }
 
-    if (log2_trafo_size > 2) {
-        if (trafo_depth == 0 ||
-            SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase)) {
-            SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
-                ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
-        }
-
-        if (trafo_depth == 0 ||
-            SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase)) {
-            SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
-                ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
-        }
-    }
+    if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cb))
+        cbf_cb = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
+    else if (log2_trafo_size > 2 || trafo_depth == 0)
+        cbf_cb = 0;
+    if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cr))
+        cbf_cr = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
+    else if (log2_trafo_size > 2 || trafo_depth == 0)
+        cbf_cr = 0;
 
     if (split_transform_flag) {
-        int x1 = x0 + ((1 << log2_trafo_size) >> 1);
-        int y1 = y0 + ((1 << log2_trafo_size) >> 1);
-
-        hls_transform_tree(s, x0, y0, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
-                           log2_trafo_size - 1, trafo_depth + 1, 0);
-        hls_transform_tree(s, x1, y0, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
-                           log2_trafo_size - 1, trafo_depth + 1, 1);
-        hls_transform_tree(s, x0, y1, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
-                           log2_trafo_size - 1, trafo_depth + 1, 2);
-        hls_transform_tree(s, x1, y1, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
-                           log2_trafo_size - 1, trafo_depth + 1, 3);
+        const int trafo_size_split = 1 << (log2_trafo_size - 1);
+        const int x1 = x0 + trafo_size_split;
+        const int y1 = y0 + trafo_size_split;
+
+#define SUBDIVIDE(x, y, idx)                                                    \
+do {                                                                            \
+    ret = hls_transform_tree(s, x, y, x0, y0, cb_xBase, cb_yBase, log2_cb_size, \
+                             log2_trafo_size - 1, trafo_depth + 1, idx,         \
+                             cbf_cb, cbf_cr);                                   \
+    if (ret < 0)                                                                \
+        return ret;                                                             \
+} while (0)
+
+        SUBDIVIDE(x0, y0, 0);
+        SUBDIVIDE(x1, y0, 1);
+        SUBDIVIDE(x0, y1, 2);
+        SUBDIVIDE(x1, y1, 3);
+
+#undef SUBDIVIDE
     } else {
-        int min_tu_size      = 1 << s->sps->log2_min_tb_size;
-        int log2_min_tu_size = s->sps->log2_min_tb_size;
-        int min_tu_width     = s->sps->min_tb_width;
+        int min_tu_size      = 1 << s->ps.sps->log2_min_tb_size;
+        int log2_min_tu_size = s->ps.sps->log2_min_tb_size;
+        int min_tu_width     = s->ps.sps->min_tb_width;
+        int cbf_luma         = 1;
 
         if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
-            SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
-            SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
-            lc->tt.cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
-        }
-
-        hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
-                           log2_cb_size, log2_trafo_size, trafo_depth, blk_idx);
+            cbf_cb || cbf_cr)
+            cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
 
+        ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
+                                 log2_cb_size, log2_trafo_size,
+                                 blk_idx, cbf_luma, cbf_cb, cbf_cr);
+        if (ret < 0)
+            return ret;
         // TODO: store cbf_luma somewhere else
-        if (lc->tt.cbf_luma) {
+        if (cbf_luma) {
             int i, j;
             for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
                 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
@@ -1379,14 +1401,13 @@ static void hls_transform_tree(HEVCContext *s, int x0, int y0,
                 }
         }
         if (!s->sh.disable_deblocking_filter_flag) {
-            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size,
-                                                  lc->slice_or_tiles_up_boundary,
-                                                  lc->slice_or_tiles_left_boundary);
-            if (s->pps->transquant_bypass_enable_flag &&
+            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size);
+            if (s->ps.pps->transquant_bypass_enable_flag &&
                 lc->cu.cu_transquant_bypass_flag)
                 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
         }
     }
+    return 0;
 }
 
 static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
@@ -1396,27 +1417,26 @@ static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
     GetBitContext gb;
     int cb_size   = 1 << log2_cb_size;
     int stride0   = s->frame->linesize[0];
-    uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
+    uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->ps.sps->pixel_shift)];
     int   stride1 = s->frame->linesize[1];
-    uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
+    uint8_t *dst1 = &s->frame->data[1][(y0 >> s->ps.sps->vshift[1]) * stride1 + ((x0 >> s->ps.sps->hshift[1]) << s->ps.sps->pixel_shift)];
     int   stride2 = s->frame->linesize[2];
-    uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
+    uint8_t *dst2 = &s->frame->data[2][(y0 >> s->ps.sps->vshift[2]) * stride2 + ((x0 >> s->ps.sps->hshift[2]) << s->ps.sps->pixel_shift)];
 
-    int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth;
-    const uint8_t *pcm = skip_bytes(&s->HEVClc.cc, (length + 7) >> 3);
+    int length         = cb_size * cb_size * s->ps.sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->ps.sps->pcm.bit_depth_chroma;
+    const uint8_t *pcm = skip_bytes(&lc->cc, (length + 7) >> 3);
     int ret;
 
-    ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
-                                          lc->slice_or_tiles_up_boundary,
-                                          lc->slice_or_tiles_left_boundary);
+    if (!s->sh.disable_deblocking_filter_flag)
+        ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
 
     ret = init_get_bits(&gb, pcm, length);
     if (ret < 0)
         return ret;
 
-    s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
-    s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth);
-    s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth);
+    s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->ps.sps->pcm.bit_depth);
+    s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->ps.sps->pcm.bit_depth_chroma);
+    s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->ps.sps->pcm.bit_depth_chroma);
     return 0;
 }
 
@@ -1464,8 +1484,8 @@ static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
     HEVCLocalContext *lc = &s->HEVClc;
     uint8_t *src         = ref->data[0];
     ptrdiff_t srcstride  = ref->linesize[0];
-    int pic_width        = s->sps->width;
-    int pic_height       = s->sps->height;
+    int pic_width        = s->ps.sps->width;
+    int pic_height       = s->ps.sps->height;
 
     int mx         = mv->x & 3;
     int my         = mv->y & 3;
@@ -1474,20 +1494,24 @@ static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
 
     x_off += mv->x >> 2;
     y_off += mv->y >> 2;
-    src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
+    src   += y_off * srcstride + (x_off * (1 << s->ps.sps->pixel_shift));
 
     if (x_off < extra_left || y_off < extra_top ||
         x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
         y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
-        int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
+        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
+        int offset = extra_top * srcstride + (extra_left << s->ps.sps->pixel_shift);
+        int buf_offset = extra_top *
+                         edge_emu_stride + (extra_left << s->ps.sps->pixel_shift);
 
         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
-                                 srcstride, srcstride,
+                                 edge_emu_stride, srcstride,
                                  block_w + ff_hevc_qpel_extra[mx],
                                  block_h + ff_hevc_qpel_extra[my],
                                  x_off - extra_left, y_off - extra_top,
                                  pic_width, pic_height);
-        src = lc->edge_emu_buffer + offset;
+        src = lc->edge_emu_buffer + buf_offset;
+        srcstride = edge_emu_stride;
     }
     s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
                                      block_h, lc->mc_buffer);
@@ -1516,41 +1540,49 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
     uint8_t *src2        = ref->data[2];
     ptrdiff_t src1stride = ref->linesize[1];
     ptrdiff_t src2stride = ref->linesize[2];
-    int pic_width        = s->sps->width >> 1;
-    int pic_height       = s->sps->height >> 1;
+    int pic_width        = s->ps.sps->width >> 1;
+    int pic_height       = s->ps.sps->height >> 1;
 
     int mx = mv->x & 7;
     int my = mv->y & 7;
 
     x_off += mv->x >> 3;
     y_off += mv->y >> 3;
-    src1  += y_off * src1stride + (x_off << s->sps->pixel_shift);
-    src2  += y_off * src2stride + (x_off << s->sps->pixel_shift);
+    src1  += y_off * src1stride + (x_off * (1 << s->ps.sps->pixel_shift));
+    src2  += y_off * src2stride + (x_off * (1 << s->ps.sps->pixel_shift));
 
     if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
         x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
         y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
-        int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
-        int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
+        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
+        int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->ps.sps->pixel_shift));
+        int buf_offset1 = EPEL_EXTRA_BEFORE *
+                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
+        int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->ps.sps->pixel_shift));
+        int buf_offset2 = EPEL_EXTRA_BEFORE *
+                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
 
         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
-                                 src1stride, src1stride,
+                                 edge_emu_stride, src1stride,
                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
                                  x_off - EPEL_EXTRA_BEFORE,
                                  y_off - EPEL_EXTRA_BEFORE,
                                  pic_width, pic_height);
 
-        src1 = lc->edge_emu_buffer + offset1;
+        src1 = lc->edge_emu_buffer + buf_offset1;
+        src1stride = edge_emu_stride;
         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
                                              block_w, block_h, mx, my, lc->mc_buffer);
 
         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
-                                 src2stride, src2stride,
+                                 edge_emu_stride, src2stride,
                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
                                  x_off - EPEL_EXTRA_BEFORE,
                                  y_off - EPEL_EXTRA_BEFORE,
                                  pic_width, pic_height);
-        src2 = lc->edge_emu_buffer + offset2;
+        src2 = lc->edge_emu_buffer + buf_offset2;
+        src2stride = edge_emu_stride;
+
         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
                                              block_w, block_h, mx, my,
                                              lc->mc_buffer);
@@ -1571,18 +1603,62 @@ static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
     ff_thread_await_progress(&ref->tf, y, 0);
 }
 
+static void hevc_luma_mv_mpv_mode(HEVCContext *s, int x0, int y0, int nPbW,
+                                  int nPbH, int log2_cb_size, int part_idx,
+                                  int merge_idx, MvField *mv)
+{
+    HEVCLocalContext *lc             = &s->HEVClc;
+    enum InterPredIdc inter_pred_idc = PRED_L0;
+    int mvp_flag;
+
+    ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
+    if (s->sh.slice_type == B_SLICE)
+        inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
+
+    if (inter_pred_idc != PRED_L1) {
+        if (s->sh.nb_refs[L0])
+            mv->ref_idx[0]= ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
+
+        mv->pred_flag[0] = 1;
+        hls_mvd_coding(s, x0, y0, 0);
+        mvp_flag = ff_hevc_mvp_lx_flag_decode(s);
+        ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                 part_idx, merge_idx, mv, mvp_flag, 0);
+        mv->mv[0].x += lc->pu.mvd.x;
+        mv->mv[0].y += lc->pu.mvd.y;
+    }
+
+    if (inter_pred_idc != PRED_L0) {
+        if (s->sh.nb_refs[L1])
+            mv->ref_idx[1]= ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
+
+        if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
+            AV_ZERO32(&lc->pu.mvd);
+        } else {
+            hls_mvd_coding(s, x0, y0, 1);
+        }
+
+        mv->pred_flag[1] = 1;
+        mvp_flag = ff_hevc_mvp_lx_flag_decode(s);
+        ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                 part_idx, merge_idx, mv, mvp_flag, 1);
+        mv->mv[1].x += lc->pu.mvd.x;
+        mv->mv[1].y += lc->pu.mvd.y;
+    }
+}
+
 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                                 int nPbW, int nPbH,
                                 int log2_cb_size, int partIdx)
 {
 #define POS(c_idx, x, y)                                                              \
-    &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
-                           (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
+    &s->frame->data[c_idx][((y) >> s->ps.sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
+                           (((x) >> s->ps.sps->hshift[c_idx]) << s->ps.sps->pixel_shift)]
     HEVCLocalContext *lc = &s->HEVClc;
     int merge_idx = 0;
     struct MvField current_mv = {{{ 0 }}};
 
-    int min_pu_width = s->sps->min_pu_width;
+    int min_pu_width = s->ps.sps->min_pu_width;
 
     MvField *tab_mvf = s->ref->tab_mvf;
     RefPicList  *refPicList = s->ref->refPicList;
@@ -1593,99 +1669,37 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
     uint8_t *dst0 = POS(0, x0, y0);
     uint8_t *dst1 = POS(1, x0, y0);
     uint8_t *dst2 = POS(2, x0, y0);
-    int log2_min_cb_size = s->sps->log2_min_cb_size;
-    int min_cb_width     = s->sps->min_cb_width;
+    int log2_min_cb_size = s->ps.sps->log2_min_cb_size;
+    int min_cb_width     = s->ps.sps->min_cb_width;
     int x_cb             = x0 >> log2_min_cb_size;
     int y_cb             = y0 >> log2_min_cb_size;
-    int ref_idx[2];
-    int mvp_flag[2];
     int x_pu, y_pu;
     int i, j;
 
-    if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
+    int skip_flag = SAMPLE_CTB(s->skip_flag, x_cb, y_cb);
+
+    if (!skip_flag)
+        lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
+
+    if (skip_flag || lc->pu.merge_flag) {
         if (s->sh.max_num_merge_cand > 1)
             merge_idx = ff_hevc_merge_idx_decode(s);
         else
             merge_idx = 0;
 
-        ff_hevc_luma_mv_merge_mode(s, x0, y0,
-                                   1 << log2_cb_size,
-                                   1 << log2_cb_size,
-                                   log2_cb_size, partIdx,
-                                   merge_idx, &current_mv);
-        x_pu = x0 >> s->sps->log2_min_pu_size;
-        y_pu = y0 >> s->sps->log2_min_pu_size;
-
-        for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
-            for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
-                tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
-    } else { /* MODE_INTER */
-        lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
-        if (lc->pu.merge_flag) {
-            if (s->sh.max_num_merge_cand > 1)
-                merge_idx = ff_hevc_merge_idx_decode(s);
-            else
-                merge_idx = 0;
-
-            ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
-                                       partIdx, merge_idx, &current_mv);
-            x_pu = x0 >> s->sps->log2_min_pu_size;
-            y_pu = y0 >> s->sps->log2_min_pu_size;
-
-            for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
-                for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
-                    tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
-        } else {
-            enum InterPredIdc inter_pred_idc = PRED_L0;
-            ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
-            if (s->sh.slice_type == B_SLICE)
-                inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
-
-            if (inter_pred_idc != PRED_L1) {
-                if (s->sh.nb_refs[L0]) {
-                    ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
-                    current_mv.ref_idx[0] = ref_idx[0];
-                }
-                current_mv.pred_flag[0] = 1;
-                hls_mvd_coding(s, x0, y0, 0);
-                mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
-                ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
-                                         partIdx, merge_idx, &current_mv,
-                                         mvp_flag[0], 0);
-                current_mv.mv[0].x += lc->pu.mvd.x;
-                current_mv.mv[0].y += lc->pu.mvd.y;
-            }
-
-            if (inter_pred_idc != PRED_L0) {
-                if (s->sh.nb_refs[L1]) {
-                    ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
-                    current_mv.ref_idx[1] = ref_idx[1];
-                }
-
-                if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
-                    lc->pu.mvd.x = 0;
-                    lc->pu.mvd.y = 0;
-                } else {
-                    hls_mvd_coding(s, x0, y0, 1);
-                }
-
-                current_mv.pred_flag[1] = 1;
-                mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
-                ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
-                                         partIdx, merge_idx, &current_mv,
-                                         mvp_flag[1], 1);
-                current_mv.mv[1].x += lc->pu.mvd.x;
-                current_mv.mv[1].y += lc->pu.mvd.y;
-            }
+        ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                   partIdx, merge_idx, &current_mv);
+    } else {
+        hevc_luma_mv_mpv_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                              partIdx, merge_idx, &current_mv);
+    }
 
-            x_pu = x0 >> s->sps->log2_min_pu_size;
-            y_pu = y0 >> s->sps->log2_min_pu_size;
+    x_pu = x0 >> s->ps.sps->log2_min_pu_size;
+    y_pu = y0 >> s->ps.sps->log2_min_pu_size;
 
-            for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
-                for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
-                    tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
-        }
-    }
+    for (j = 0; j < nPbH >> s->ps.sps->log2_min_pu_size; j++)
+        for (i = 0; i < nPbW >> s->ps.sps->log2_min_pu_size; i++)
+            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
 
     if (current_mv.pred_flag[0]) {
         ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
@@ -1707,8 +1721,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         luma_mc(s, tmp, tmpstride, ref0->frame,
                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
+            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
             s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
                                      s->sh.luma_weight_l0[current_mv.ref_idx[0]],
                                      s->sh.luma_offset_l0[current_mv.ref_idx[0]],
@@ -1720,8 +1734,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
                   &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
+            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
                                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
                                      s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
@@ -1740,14 +1754,11 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
 
-        if (!ref1)
-            return;
-
         luma_mc(s, tmp, tmpstride, ref1->frame,
                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
+            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
             s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
                                       s->sh.luma_weight_l1[current_mv.ref_idx[1]],
                                       s->sh.luma_offset_l1[current_mv.ref_idx[1]],
@@ -1760,8 +1771,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
                   &current_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
+            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
                                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
                                      s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
@@ -1779,19 +1790,14 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
         DECLARE_ALIGNED(16, int16_t, tmp3[MAX_PB_SIZE * MAX_PB_SIZE]);
         DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
-        HEVCFrame *ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
-        HEVCFrame *ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
-
-        if (!ref0 || !ref1)
-            return;
 
         luma_mc(s, tmp, tmpstride, ref0->frame,
                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
         luma_mc(s, tmp2, tmpstride, ref1->frame,
                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
+            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
             s->hevcdsp.weighted_pred_avg(s->sh.luma_log2_weight_denom,
                                          s->sh.luma_weight_l0[current_mv.ref_idx[0]],
                                          s->sh.luma_weight_l1[current_mv.ref_idx[1]],
@@ -1800,8 +1806,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                                          dst0, s->frame->linesize[0],
                                          tmp, tmp2, tmpstride, nPbW, nPbH);
         } else {
-            s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
-                                             tmp, tmp2, tmpstride, nPbW, nPbH);
+            s->hevcdsp.put_unweighted_pred_avg(dst0, s->frame->linesize[0],
+                                               tmp, tmp2, tmpstride, nPbW, nPbH);
         }
 
         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
@@ -1809,8 +1815,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
                   &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
+            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
             s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
                                          s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
                                          s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
@@ -1826,8 +1832,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                                          dst2, s->frame->linesize[2], tmp2, tmp4,
                                          tmpstride, nPbW / 2, nPbH / 2);
         } else {
-            s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
-            s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
+            s->hevcdsp.put_unweighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
+            s->hevcdsp.put_unweighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
         }
     }
 }
@@ -1839,19 +1845,19 @@ static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
                                 int prev_intra_luma_pred_flag)
 {
     HEVCLocalContext *lc = &s->HEVClc;
-    int x_pu             = x0 >> s->sps->log2_min_pu_size;
-    int y_pu             = y0 >> s->sps->log2_min_pu_size;
-    int min_pu_width     = s->sps->min_pu_width;
-    int size_in_pus      = pu_size >> s->sps->log2_min_pu_size;
-    int x0b              = x0 & ((1 << s->sps->log2_ctb_size) - 1);
-    int y0b              = y0 & ((1 << s->sps->log2_ctb_size) - 1);
+    int x_pu             = x0 >> s->ps.sps->log2_min_pu_size;
+    int y_pu             = y0 >> s->ps.sps->log2_min_pu_size;
+    int min_pu_width     = s->ps.sps->min_pu_width;
+    int size_in_pus      = pu_size >> s->ps.sps->log2_min_pu_size;
+    int x0b              = x0 & ((1 << s->ps.sps->log2_ctb_size) - 1);
+    int y0b              = y0 & ((1 << s->ps.sps->log2_ctb_size) - 1);
 
     int cand_up   = (lc->ctb_up_flag || y0b) ?
                     s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
     int cand_left = (lc->ctb_left_flag || x0b) ?
                     s->tab_ipm[y_pu * min_pu_width + x_pu - 1]   : INTRA_DC;
 
-    int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
+    int y_ctb = (y0 >> (s->ps.sps->log2_ctb_size)) << (s->ps.sps->log2_ctb_size);
 
     MvField *tab_mvf = s->ref->tab_mvf;
     int intra_pred_mode;
@@ -1926,13 +1932,13 @@ static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
                                           int log2_cb_size, int ct_depth)
 {
-    int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
-    int x_cb   = x0 >> s->sps->log2_min_cb_size;
-    int y_cb   = y0 >> s->sps->log2_min_cb_size;
+    int length = (1 << log2_cb_size) >> s->ps.sps->log2_min_cb_size;
+    int x_cb   = x0 >> s->ps.sps->log2_min_cb_size;
+    int y_cb   = y0 >> s->ps.sps->log2_min_cb_size;
     int y;
 
     for (y = 0; y < length; y++)
-        memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
+        memset(&s->tab_ct_depth[(y_cb + y) * s->ps.sps->min_cb_width + x_cb],
                ct_depth, length);
 }
 
@@ -1982,11 +1988,11 @@ static void intra_prediction_unit_default_value(HEVCContext *s,
 {
     HEVCLocalContext *lc = &s->HEVClc;
     int pb_size          = 1 << log2_cb_size;
-    int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
-    int min_pu_width     = s->sps->min_pu_width;
+    int size_in_pus      = pb_size >> s->ps.sps->log2_min_pu_size;
+    int min_pu_width     = s->ps.sps->min_pu_width;
     MvField *tab_mvf     = s->ref->tab_mvf;
-    int x_pu             = x0 >> s->sps->log2_min_pu_size;
-    int y_pu             = y0 >> s->sps->log2_min_pu_size;
+    int x_pu             = x0 >> s->ps.sps->log2_min_pu_size;
+    int y_pu             = y0 >> s->ps.sps->log2_min_pu_size;
     int j, k;
 
     if (size_in_pus == 0)
@@ -2002,25 +2008,23 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
 {
     int cb_size          = 1 << log2_cb_size;
     HEVCLocalContext *lc = &s->HEVClc;
-    int log2_min_cb_size = s->sps->log2_min_cb_size;
+    int log2_min_cb_size = s->ps.sps->log2_min_cb_size;
     int length           = cb_size >> log2_min_cb_size;
-    int min_cb_width     = s->sps->min_cb_width;
+    int min_cb_width     = s->ps.sps->min_cb_width;
     int x_cb             = x0 >> log2_min_cb_size;
     int y_cb             = y0 >> log2_min_cb_size;
-    int x, y;
+    int x, y, ret;
 
     lc->cu.x                = x0;
     lc->cu.y                = y0;
-    lc->cu.rqt_root_cbf     = 1;
     lc->cu.pred_mode        = MODE_INTRA;
     lc->cu.part_mode        = PART_2Nx2N;
     lc->cu.intra_split_flag = 0;
-    lc->cu.pcm_flag         = 0;
 
     SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
     for (x = 0; x < 4; x++)
         lc->pu.intra_pred_mode[x] = 1;
-    if (s->pps->transquant_bypass_enable_flag) {
+    if (s->ps.pps->transquant_bypass_enable_flag) {
         lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
         if (lc->cu.cu_transquant_bypass_flag)
             set_deblocking_bypass(s, x0, y0, log2_cb_size);
@@ -2030,7 +2034,6 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
     if (s->sh.slice_type != I_SLICE) {
         uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
 
-        lc->cu.pred_mode = MODE_SKIP;
         x = y_cb * min_cb_width + x_cb;
         for (y = 0; y < length; y++) {
             memset(&s->skip_flag[x], skip_flag, length);
@@ -2044,30 +2047,29 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
         intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
 
         if (!s->sh.disable_deblocking_filter_flag)
-            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
-                                                  lc->slice_or_tiles_up_boundary,
-                                                  lc->slice_or_tiles_left_boundary);
+            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
     } else {
+        int pcm_flag = 0;
+
         if (s->sh.slice_type != I_SLICE)
             lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
         if (lc->cu.pred_mode != MODE_INTRA ||
-            log2_cb_size == s->sps->log2_min_cb_size) {
+            log2_cb_size == s->ps.sps->log2_min_cb_size) {
             lc->cu.part_mode        = ff_hevc_part_mode_decode(s, log2_cb_size);
             lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
                                       lc->cu.pred_mode == MODE_INTRA;
         }
 
         if (lc->cu.pred_mode == MODE_INTRA) {
-            if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
-                log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
-                log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
-                lc->cu.pcm_flag = ff_hevc_pcm_flag_decode(s);
+            if (lc->cu.part_mode == PART_2Nx2N && s->ps.sps->pcm_enabled_flag &&
+                log2_cb_size >= s->ps.sps->pcm.log2_min_pcm_cb_size &&
+                log2_cb_size <= s->ps.sps->pcm.log2_max_pcm_cb_size) {
+                pcm_flag = ff_hevc_pcm_flag_decode(s);
             }
-            if (lc->cu.pcm_flag) {
-                int ret;
+            if (pcm_flag) {
                 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
                 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
-                if (s->sps->pcm.loop_filter_disable_flag)
+                if (s->ps.sps->pcm.loop_filter_disable_flag)
                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
 
                 if (ret < 0)
@@ -2114,27 +2116,30 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
             }
         }
 
-        if (!lc->cu.pcm_flag) {
+        if (!pcm_flag) {
+            int rqt_root_cbf = 1;
+
             if (lc->cu.pred_mode != MODE_INTRA &&
                 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
-                lc->cu.rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
+                rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
             }
-            if (lc->cu.rqt_root_cbf) {
+            if (rqt_root_cbf) {
                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
-                                         s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
-                                         s->sps->max_transform_hierarchy_depth_inter;
-                hls_transform_tree(s, x0, y0, x0, y0, x0, y0, log2_cb_size,
-                                   log2_cb_size, 0, 0);
+                                         s->ps.sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
+                                         s->ps.sps->max_transform_hierarchy_depth_inter;
+                ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
+                                         log2_cb_size,
+                                         log2_cb_size, 0, 0, 0, 0);
+                if (ret < 0)
+                    return ret;
             } else {
                 if (!s->sh.disable_deblocking_filter_flag)
-                    ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
-                                                          lc->slice_or_tiles_up_boundary,
-                                                          lc->slice_or_tiles_left_boundary);
+                    ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
             }
         }
     }
 
-    if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
+    if (s->ps.pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
         ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
 
     x = y_cb * min_cb_width + x_cb;
@@ -2153,24 +2158,23 @@ static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
 {
     HEVCLocalContext *lc = &s->HEVClc;
     const int cb_size    = 1 << log2_cb_size;
+    int split_cu;
 
     lc->ct.depth = cb_depth;
-    if (x0 + cb_size <= s->sps->width  &&
-        y0 + cb_size <= s->sps->height &&
-        log2_cb_size > s->sps->log2_min_cb_size) {
-        SAMPLE(s->split_cu_flag, x0, y0) =
-            ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
+    if (x0 + cb_size <= s->ps.sps->width  &&
+        y0 + cb_size <= s->ps.sps->height &&
+        log2_cb_size > s->ps.sps->log2_min_cb_size) {
+        split_cu = ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
     } else {
-        SAMPLE(s->split_cu_flag, x0, y0) =
-            (log2_cb_size > s->sps->log2_min_cb_size);
+        split_cu = (log2_cb_size > s->ps.sps->log2_min_cb_size);
     }
-    if (s->pps->cu_qp_delta_enabled_flag &&
-        log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
+    if (s->ps.pps->cu_qp_delta_enabled_flag &&
+        log2_cb_size >= s->ps.sps->log2_ctb_size - s->ps.pps->diff_cu_qp_delta_depth) {
         lc->tu.is_cu_qp_delta_coded = 0;
         lc->tu.cu_qp_delta          = 0;
     }
 
-    if (SAMPLE(s->split_cu_flag, x0, y0)) {
+    if (split_cu) {
         const int cb_size_split = cb_size >> 1;
         const int x1 = x0 + cb_size_split;
         const int y1 = y0 + cb_size_split;
@@ -2180,7 +2184,7 @@ static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
 
 #define SUBDIVIDE(x, y)                                                \
 do {                                                                   \
-    if (x < s->sps->width && y < s->sps->height) {                     \
+    if (x < s->ps.sps->width && y < s->ps.sps->height) {                     \
         int ret = hls_coding_quadtree(s, x, y, log2_cb_size, cb_depth);\
         if (ret < 0)                                                   \
             return ret;                                                \
@@ -2204,80 +2208,77 @@ static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
                                  int ctb_addr_ts)
 {
     HEVCLocalContext *lc  = &s->HEVClc;
-    int ctb_size          = 1 << s->sps->log2_ctb_size;
-    int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
+    int ctb_size          = 1 << s->ps.sps->log2_ctb_size;
+    int ctb_addr_rs       = s->ps.pps->ctb_addr_ts_to_rs[ctb_addr_ts];
     int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
 
-    int tile_left_boundary, tile_up_boundary;
-    int slice_left_boundary, slice_up_boundary;
-
     s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
 
-    if (s->pps->entropy_coding_sync_enabled_flag) {
+    if (s->ps.pps->entropy_coding_sync_enabled_flag) {
         if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
             lc->first_qp_group = 1;
-        lc->end_of_tiles_x = s->sps->width;
-    } else if (s->pps->tiles_enabled_flag) {
-        if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
-            int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
+        lc->end_of_tiles_x = s->ps.sps->width;
+    } else if (s->ps.pps->tiles_enabled_flag) {
+        if (ctb_addr_ts && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[ctb_addr_ts - 1]) {
+            int idxX = s->ps.pps->col_idxX[x_ctb >> s->ps.sps->log2_ctb_size];
             lc->start_of_tiles_x = x_ctb;
-            lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
+            lc->end_of_tiles_x   = x_ctb + (s->ps.pps->column_width[idxX] << s->ps.sps->log2_ctb_size);
             lc->first_qp_group   = 1;
         }
     } else {
-        lc->end_of_tiles_x = s->sps->width;
+        lc->end_of_tiles_x = s->ps.sps->width;
     }
 
-    lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
+    lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->ps.sps->height);
 
-    if (s->pps->tiles_enabled_flag) {
-        tile_left_boundary  = x_ctb > 0 &&
-                              s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - 1]];
-        slice_left_boundary = x_ctb > 0 &&
-                              s->tab_slice_address[ctb_addr_rs] == s->tab_slice_address[ctb_addr_rs - 1];
-        tile_up_boundary  = y_ctb > 0 &&
-                            s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
-        slice_up_boundary = y_ctb > 0 &&
-                            s->tab_slice_address[ctb_addr_rs] == s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
+    lc->boundary_flags = 0;
+    if (s->ps.pps->tiles_enabled_flag) {
+        if (x_ctb > 0 && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs - 1]])
+            lc->boundary_flags |= BOUNDARY_LEFT_TILE;
+        if (x_ctb > 0 && s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1])
+            lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
+        if (y_ctb > 0 && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->ps.sps->ctb_width]])
+            lc->boundary_flags |= BOUNDARY_UPPER_TILE;
+        if (y_ctb > 0 && s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->ps.sps->ctb_width])
+            lc->boundary_flags |= BOUNDARY_UPPER_SLICE;
     } else {
-        tile_left_boundary  =
-        tile_up_boundary    = 1;
-        slice_left_boundary = ctb_addr_in_slice > 0;
-        slice_up_boundary   = ctb_addr_in_slice >= s->sps->ctb_width;
-    }
-    lc->slice_or_tiles_left_boundary = (!slice_left_boundary) + (!tile_left_boundary << 1);
-    lc->slice_or_tiles_up_boundary   = (!slice_up_boundary + (!tile_up_boundary << 1));
-    lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0) && tile_left_boundary);
-    lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && tile_up_boundary);
-    lc->ctb_up_right_flag = ((y_ctb > 0)  && (ctb_addr_in_slice+1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->sps->ctb_width]]));
-    lc->ctb_up_left_flag = ((x_ctb > 0) && (y_ctb > 0)  && (ctb_addr_in_slice-1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1 - s->sps->ctb_width]]));
+        if (!ctb_addr_in_slice > 0)
+            lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
+        if (ctb_addr_in_slice < s->ps.sps->ctb_width)
+            lc->boundary_flags |= BOUNDARY_UPPER_SLICE;
+    }
+
+    lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0) && !(lc->boundary_flags & BOUNDARY_LEFT_TILE));
+    lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->ps.sps->ctb_width) && !(lc->boundary_flags & BOUNDARY_UPPER_TILE));
+    lc->ctb_up_right_flag = ((y_ctb > 0)  && (ctb_addr_in_slice+1 >= s->ps.sps->ctb_width) && (s->ps.pps->tile_id[ctb_addr_ts] == s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->ps.sps->ctb_width]]));
+    lc->ctb_up_left_flag = ((x_ctb > 0) && (y_ctb > 0)  && (ctb_addr_in_slice-1 >= s->ps.sps->ctb_width) && (s->ps.pps->tile_id[ctb_addr_ts] == s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs-1 - s->ps.sps->ctb_width]]));
 }
 
 static int hls_slice_data(HEVCContext *s)
 {
-    int ctb_size    = 1 << s->sps->log2_ctb_size;
+    int ctb_size    = 1 << s->ps.sps->log2_ctb_size;
     int more_data   = 1;
     int x_ctb       = 0;
     int y_ctb       = 0;
-    int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
+    int ctb_addr_ts = s->ps.pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
     int ret;
 
-    while (more_data && ctb_addr_ts < s->sps->ctb_size) {
-        int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
+    while (more_data && ctb_addr_ts < s->ps.sps->ctb_size) {
+        int ctb_addr_rs = s->ps.pps->ctb_addr_ts_to_rs[ctb_addr_ts];
 
-        x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
-        y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
+        x_ctb = (ctb_addr_rs % ((s->ps.sps->width + ctb_size - 1) >> s->ps.sps->log2_ctb_size)) << s->ps.sps->log2_ctb_size;
+        y_ctb = (ctb_addr_rs / ((s->ps.sps->width + ctb_size - 1) >> s->ps.sps->log2_ctb_size)) << s->ps.sps->log2_ctb_size;
         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
 
         ff_hevc_cabac_init(s, ctb_addr_ts);
 
-        hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
+        hls_sao_param(s, x_ctb >> s->ps.sps->log2_ctb_size, y_ctb >> s->ps.sps->log2_ctb_size);
 
         s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
         s->deblock[ctb_addr_rs].tc_offset   = s->sh.tc_offset;
         s->filter_slice_edges[ctb_addr_rs]  = s->sh.slice_loop_filter_across_slices_enabled_flag;
 
-        ret = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
+        ret = hls_coding_quadtree(s, x_ctb, y_ctb, s->ps.sps->log2_ctb_size, 0);
         if (ret < 0)
             return ret;
         more_data = !ff_hevc_end_of_slice_flag_decode(s);
@@ -2287,55 +2288,29 @@ static int hls_slice_data(HEVCContext *s)
         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
     }
 
-    if (x_ctb + ctb_size >= s->sps->width &&
-        y_ctb + ctb_size >= s->sps->height)
+    if (x_ctb + ctb_size >= s->ps.sps->width &&
+        y_ctb + ctb_size >= s->ps.sps->height)
         ff_hevc_hls_filter(s, x_ctb, y_ctb);
 
     return ctb_addr_ts;
 }
 
-/**
- * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
- * 0 if the unit should be skipped, 1 otherwise
- */
-static int hls_nal_unit(HEVCContext *s)
-{
-    GetBitContext *gb = &s->HEVClc.gb;
-    int nuh_layer_id;
-
-    if (get_bits1(gb) != 0)
-        return AVERROR_INVALIDDATA;
-
-    s->nal_unit_type = get_bits(gb, 6);
-
-    nuh_layer_id   = get_bits(gb, 6);
-    s->temporal_id = get_bits(gb, 3) - 1;
-    if (s->temporal_id < 0)
-        return AVERROR_INVALIDDATA;
-
-    av_log(s->avctx, AV_LOG_DEBUG,
-           "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
-           s->nal_unit_type, nuh_layer_id, s->temporal_id);
-
-    return nuh_layer_id == 0;
-}
-
 static void restore_tqb_pixels(HEVCContext *s)
 {
-    int min_pu_size = 1 << s->sps->log2_min_pu_size;
+    int min_pu_size = 1 << s->ps.sps->log2_min_pu_size;
     int x, y, c_idx;
 
     for (c_idx = 0; c_idx < 3; c_idx++) {
         ptrdiff_t stride = s->frame->linesize[c_idx];
-        int hshift       = s->sps->hshift[c_idx];
-        int vshift       = s->sps->vshift[c_idx];
-        for (y = 0; y < s->sps->min_pu_height; y++) {
-            for (x = 0; x < s->sps->min_pu_width; x++) {
-                if (s->is_pcm[y * s->sps->min_pu_width + x]) {
+        int hshift       = s->ps.sps->hshift[c_idx];
+        int vshift       = s->ps.sps->vshift[c_idx];
+        for (y = 0; y < s->ps.sps->min_pu_height; y++) {
+            for (x = 0; x < s->ps.sps->min_pu_width; x++) {
+                if (s->is_pcm[y * s->ps.sps->min_pu_width + x]) {
                     int n;
                     int len      = min_pu_size >> hshift;
-                    uint8_t *src = &s->frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
-                    uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
+                    uint8_t *src = &s->frame->data[c_idx][((y << s->ps.sps->log2_min_pu_size) >> vshift) * stride + (((x << s->ps.sps->log2_min_pu_size) >> hshift) << s->ps.sps->pixel_shift)];
+                    uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->ps.sps->log2_min_pu_size) >> vshift) * stride + (((x << s->ps.sps->log2_min_pu_size) >> hshift) << s->ps.sps->pixel_shift)];
                     for (n = 0; n < (min_pu_size >> vshift); n++) {
                         memcpy(dst, src, len);
                         src += stride;
@@ -2347,6 +2322,55 @@ static void restore_tqb_pixels(HEVCContext *s)
     }
 }
 
+static int set_side_data(HEVCContext *s)
+{
+    AVFrame *out = s->ref->frame;
+
+    if (s->sei_frame_packing_present &&
+        s->frame_packing_arrangement_type >= 3 &&
+        s->frame_packing_arrangement_type <= 5 &&
+        s->content_interpretation_type > 0 &&
+        s->content_interpretation_type < 3) {
+        AVStereo3D *stereo = av_stereo3d_create_side_data(out);
+        if (!stereo)
+            return AVERROR(ENOMEM);
+
+        switch (s->frame_packing_arrangement_type) {
+        case 3:
+            if (s->quincunx_subsampling)
+                stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
+            else
+                stereo->type = AV_STEREO3D_SIDEBYSIDE;
+            break;
+        case 4:
+            stereo->type = AV_STEREO3D_TOPBOTTOM;
+            break;
+        case 5:
+            stereo->type = AV_STEREO3D_FRAMESEQUENCE;
+            break;
+        }
+
+        if (s->content_interpretation_type == 2)
+            stereo->flags = AV_STEREO3D_FLAG_INVERT;
+    }
+
+    if (s->sei_display_orientation_present &&
+        (s->sei_anticlockwise_rotation || s->sei_hflip || s->sei_vflip)) {
+        double angle = s->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
+        AVFrameSideData *rotation = av_frame_new_side_data(out,
+                                                           AV_FRAME_DATA_DISPLAYMATRIX,
+                                                           sizeof(int32_t) * 9);
+        if (!rotation)
+            return AVERROR(ENOMEM);
+
+        av_display_rotation_set((int32_t *)rotation->data, angle);
+        av_display_matrix_flip((int32_t *)rotation->data,
+                               s->sei_hflip, s->sei_vflip);
+    }
+
+    return 0;
+}
+
 static int hevc_frame_start(HEVCContext *s)
 {
     HEVCLocalContext *lc = &s->HEVClc;
@@ -2354,33 +2378,33 @@ static int hevc_frame_start(HEVCContext *s)
 
     memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
     memset(s->vertical_bs,   0, 2 * s->bs_width * (s->bs_height + 1));
-    memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
-    memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
+    memset(s->cbf_luma,      0, s->ps.sps->min_tb_width * s->ps.sps->min_tb_height);
+    memset(s->is_pcm,        0, s->ps.sps->min_pu_width * s->ps.sps->min_pu_height);
 
     lc->start_of_tiles_x = 0;
     s->is_decoded        = 0;
+    s->first_nal_type    = s->nal_unit_type;
 
-    if (s->pps->tiles_enabled_flag)
-        lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
+    if (s->ps.pps->tiles_enabled_flag)
+        lc->end_of_tiles_x = s->ps.pps->column_width[0] << s->ps.sps->log2_ctb_size;
 
-    ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
+    ret = ff_hevc_set_new_ref(s, s->ps.sps->sao_enabled ? &s->sao_frame : &s->frame,
                               s->poc);
     if (ret < 0)
         goto fail;
 
-    av_fast_malloc(&lc->edge_emu_buffer, &lc->edge_emu_buffer_size,
-                   (MAX_PB_SIZE + 7) * s->ref->frame->linesize[0]);
-    if (!lc->edge_emu_buffer) {
-        ret = AVERROR(ENOMEM);
-        goto fail;
-    }
-
     ret = ff_hevc_frame_rps(s);
     if (ret < 0) {
         av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
         goto fail;
     }
 
+    s->ref->frame->key_frame = IS_IRAP(s);
+
+    ret = set_side_data(s);
+    if (ret < 0)
+        goto fail;
+
     av_frame_unref(s->output_frame);
     ret = ff_hevc_output_frame(s, s->output_frame, 0);
     if (ret < 0)
@@ -2392,52 +2416,43 @@ static int hevc_frame_start(HEVCContext *s)
 
 fail:
     if (s->ref)
-        ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
+        ff_hevc_unref_frame(s, s->ref, ~0);
     s->ref = NULL;
     return ret;
 }
 
-static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
+static int decode_nal_unit(HEVCContext *s, const HEVCNAL *nal)
 {
     HEVCLocalContext *lc = &s->HEVClc;
     GetBitContext *gb    = &lc->gb;
     int ctb_addr_ts, ret;
 
-    ret = init_get_bits8(gb, nal, length);
-    if (ret < 0)
-        return ret;
-
-    ret = hls_nal_unit(s);
-    if (ret < 0) {
-        av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
-               s->nal_unit_type);
-        if (s->avctx->err_recognition & AV_EF_EXPLODE)
-            return ret;
-        return 0;
-    } else if (!ret)
-        return 0;
+    *gb              = nal->gb;
+    s->nal_unit_type = nal->type;
+    s->temporal_id   = nal->temporal_id;
 
     switch (s->nal_unit_type) {
     case NAL_VPS:
-        ret = ff_hevc_decode_nal_vps(s);
+        ret = ff_hevc_decode_nal_vps(gb, s->avctx, &s->ps);
         if (ret < 0)
-            return ret;
+            goto fail;
         break;
     case NAL_SPS:
-        ret = ff_hevc_decode_nal_sps(s);
+        ret = ff_hevc_decode_nal_sps(gb, s->avctx, &s->ps,
+                                     s->apply_defdispwin);
         if (ret < 0)
-            return ret;
+            goto fail;
         break;
     case NAL_PPS:
-        ret = ff_hevc_decode_nal_pps(s);
+        ret = ff_hevc_decode_nal_pps(gb, s->avctx, &s->ps);
         if (ret < 0)
-            return ret;
+            goto fail;
         break;
     case NAL_SEI_PREFIX:
     case NAL_SEI_SUFFIX:
         ret = ff_hevc_decode_nal_sei(s);
         if (ret < 0)
-            return ret;
+            goto fail;
         break;
     case NAL_TRAIL_R:
     case NAL_TRAIL_N:
@@ -2483,6 +2498,13 @@ static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
                 return ret;
         } else if (!s->ref) {
             av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
+            goto fail;
+        }
+
+        if (s->nal_unit_type != s->first_nal_type) {
+            av_log(s->avctx, AV_LOG_ERROR,
+                   "Non-matching NAL types of the VCL NALUs: %d %d\n",
+                   s->first_nal_type, s->nal_unit_type);
             return AVERROR_INVALIDDATA;
         }
 
@@ -2492,22 +2514,35 @@ static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
             if (ret < 0) {
                 av_log(s->avctx, AV_LOG_WARNING,
                        "Error constructing the reference lists for the current slice.\n");
-                if (s->avctx->err_recognition & AV_EF_EXPLODE)
-                    return ret;
+                goto fail;
             }
         }
 
-        ctb_addr_ts = hls_slice_data(s);
-        if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
-            s->is_decoded = 1;
-            if ((s->pps->transquant_bypass_enable_flag ||
-                 (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
-                s->sps->sao_enabled)
-                restore_tqb_pixels(s);
+        if (s->sh.first_slice_in_pic_flag && s->avctx->hwaccel) {
+            ret = s->avctx->hwaccel->start_frame(s->avctx, NULL, 0);
+            if (ret < 0)
+                goto fail;
         }
 
-        if (ctb_addr_ts < 0)
-            return ctb_addr_ts;
+        if (s->avctx->hwaccel) {
+            ret = s->avctx->hwaccel->decode_slice(s->avctx, nal->raw_data, nal->raw_size);
+            if (ret < 0)
+                goto fail;
+        } else {
+            ctb_addr_ts = hls_slice_data(s);
+            if (ctb_addr_ts >= (s->ps.sps->ctb_width * s->ps.sps->ctb_height)) {
+                s->is_decoded = 1;
+                if ((s->ps.pps->transquant_bypass_enable_flag ||
+                     (s->ps.sps->pcm.loop_filter_disable_flag && s->ps.sps->pcm_enabled_flag)) &&
+                    s->ps.sps->sao_enabled)
+                    restore_tqb_pixels(s);
+            }
+
+            if (ctb_addr_ts < 0) {
+                ret = ctb_addr_ts;
+                goto fail;
+            }
+        }
         break;
     case NAL_EOS_NUT:
     case NAL_EOB_NUT:
@@ -2523,188 +2558,42 @@ static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
     }
 
     return 0;
-}
-
-/* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
- * between these functions would be nice. */
-static int extract_rbsp(const uint8_t *src, int length,
-                        HEVCNAL *nal)
-{
-    int i, si, di;
-    uint8_t *dst;
-
-#define STARTCODE_TEST                                                  \
-        if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
-            if (src[i + 2] != 3) {                                      \
-                /* startcode, so we must be past the end */             \
-                length = i;                                             \
-            }                                                           \
-            break;                                                      \
-        }
-#if HAVE_FAST_UNALIGNED
-#define FIND_FIRST_ZERO                                                 \
-        if (i > 0 && !src[i])                                           \
-            i--;                                                        \
-        while (src[i])                                                  \
-            i++
-#if HAVE_FAST_64BIT
-    for (i = 0; i + 1 < length; i += 9) {
-        if (!((~AV_RN64A(src + i) &
-               (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
-              0x8000800080008080ULL))
-            continue;
-        FIND_FIRST_ZERO;
-        STARTCODE_TEST;
-        i -= 7;
-    }
-#else
-    for (i = 0; i + 1 < length; i += 5) {
-        if (!((~AV_RN32A(src + i) &
-               (AV_RN32A(src + i) - 0x01000101U)) &
-              0x80008080U))
-            continue;
-        FIND_FIRST_ZERO;
-        STARTCODE_TEST;
-        i -= 3;
-    }
-#endif /* HAVE_FAST_64BIT */
-#else
-    for (i = 0; i + 1 < length; i += 2) {
-        if (src[i])
-            continue;
-        if (i > 0 && src[i - 1] == 0)
-            i--;
-        STARTCODE_TEST;
-    }
-#endif /* HAVE_FAST_UNALIGNED */
-
-    if (i >= length - 1) { // no escaped 0
-        nal->data = src;
-        nal->size = length;
-        return length;
-    }
-
-    av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
-                   length + FF_INPUT_BUFFER_PADDING_SIZE);
-    if (!nal->rbsp_buffer)
-        return AVERROR(ENOMEM);
-
-    dst = nal->rbsp_buffer;
-
-    memcpy(dst, src, i);
-    si = di = i;
-    while (si + 2 < length) {
-        // remove escapes (very rare 1:2^22)
-        if (src[si + 2] > 3) {
-            dst[di++] = src[si++];
-            dst[di++] = src[si++];
-        } else if (src[si] == 0 && src[si + 1] == 0) {
-            if (src[si + 2] == 3) { // escape
-                dst[di++] = 0;
-                dst[di++] = 0;
-                si       += 3;
-
-                continue;
-            } else // next start code
-                goto nsc;
-        }
-
-        dst[di++] = src[si++];
-    }
-    while (si < length)
-        dst[di++] = src[si++];
-
-nsc:
-    memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
-
-    nal->data = dst;
-    nal->size = di;
-    return si;
+fail:
+    if (s->avctx->err_recognition & AV_EF_EXPLODE)
+        return ret;
+    return 0;
 }
 
 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
 {
-    int i, consumed, ret = 0;
+    int i, ret = 0;
 
     s->ref = NULL;
     s->eos = 0;
 
     /* split the input packet into NAL units, so we know the upper bound on the
      * number of slices in the frame */
-    s->nb_nals = 0;
-    while (length >= 4) {
-        HEVCNAL *nal;
-        int extract_length = 0;
-
-        if (s->is_nalff) {
-            int i;
-            for (i = 0; i < s->nal_length_size; i++)
-                extract_length = (extract_length << 8) | buf[i];
-            buf    += s->nal_length_size;
-            length -= s->nal_length_size;
-
-            if (extract_length > length) {
-                av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
-                ret = AVERROR_INVALIDDATA;
-                goto fail;
-            }
-        } else {
-            if (buf[2] == 0) {
-                length--;
-                buf++;
-                continue;
-            }
-            if (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
-                ret = AVERROR_INVALIDDATA;
-                goto fail;
-            }
-
-            buf           += 3;
-            length        -= 3;
-            extract_length = length;
-        }
-
-        if (s->nals_allocated < s->nb_nals + 1) {
-            int new_size = s->nals_allocated + 1;
-            HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
-            if (!tmp) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
-            s->nals = tmp;
-            memset(s->nals + s->nals_allocated, 0,
-                   (new_size - s->nals_allocated) * sizeof(*tmp));
-            s->nals_allocated = new_size;
-        }
-        nal = &s->nals[s->nb_nals++];
-
-        consumed = extract_rbsp(buf, extract_length, nal);
-        if (consumed < 0) {
-            ret = consumed;
-            goto fail;
-        }
-
-        ret = init_get_bits8(&s->HEVClc.gb, nal->data, nal->size);
-        if (ret < 0)
-            goto fail;
-        hls_nal_unit(s);
+    ret = ff_hevc_split_packet(&s->pkt, buf, length, s->avctx, s->is_nalff,
+                               s->nal_length_size);
+    if (ret < 0) {
+        av_log(s->avctx, AV_LOG_ERROR,
+               "Error splitting the input into NAL units.\n");
+        return ret;
+    }
 
-        if (s->nal_unit_type == NAL_EOB_NUT ||
-            s->nal_unit_type == NAL_EOS_NUT)
+    for (i = 0; i < s->pkt.nb_nals; i++) {
+        if (s->pkt.nals[i].type == NAL_EOB_NUT ||
+            s->pkt.nals[i].type == NAL_EOS_NUT)
             s->eos = 1;
-
-        buf    += consumed;
-        length -= consumed;
     }
 
-    /* parse the NAL units */
-    for (i = 0; i < s->nb_nals; i++) {
-        int ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
+    /* decode the NAL units */
+    for (i = 0; i < s->pkt.nb_nals; i++) {
+        ret = decode_nal_unit(s, &s->pkt.nals[i]);
         if (ret < 0) {
             av_log(s->avctx, AV_LOG_WARNING,
                    "Error parsing NAL unit #%d.\n", i);
-            if (s->avctx->err_recognition & AV_EF_EXPLODE)
-                goto fail;
+            goto fail;
         }
     }
 
@@ -2725,12 +2614,14 @@ static void print_md5(void *log_ctx, int level, uint8_t md5[16])
 static int verify_md5(HEVCContext *s, AVFrame *frame)
 {
     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
-    int pixel_shift = desc->comp[0].depth_minus1 > 7;
+    int pixel_shift;
     int i, j;
 
     if (!desc)
         return AVERROR(EINVAL);
 
+    pixel_shift = desc->comp[0].depth > 8;
+
     av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
            s->poc);
 
@@ -2758,8 +2649,8 @@ static int verify_md5(HEVCContext *s, AVFrame *frame)
             const uint8_t *src = frame->data[i] + j * frame->linesize[i];
 #if HAVE_BIGENDIAN
             if (pixel_shift) {
-                s->dsp.bswap16_buf((uint16_t*)s->checksum_buf,
-                                   (const uint16_t*)src, w);
+                s->bdsp.bswap16_buf((uint16_t *) s->checksum_buf,
+                                    (const uint16_t *) src, w);
                 src = s->checksum_buf;
             }
 #endif
@@ -2806,13 +2697,19 @@ static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
     if (ret < 0)
         return ret;
 
-    /* verify the SEI checksum */
-    if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
-        s->is_md5) {
-        ret = verify_md5(s, s->ref->frame);
-        if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
-            ff_hevc_unref_frame(s, s->ref, ~0);
-            return ret;
+    if (avctx->hwaccel) {
+        if (s->ref && avctx->hwaccel->end_frame(avctx) < 0)
+            av_log(avctx, AV_LOG_ERROR,
+                   "hardware accelerator failed to decode picture\n");
+    } else {
+        /* verify the SEI checksum */
+        if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
+            s->is_md5) {
+            ret = verify_md5(s, s->ref->frame);
+            if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
+                ff_hevc_unref_frame(s, s->ref, ~0);
+                return ret;
+            }
         }
     }
     s->is_md5 = 0;
@@ -2856,6 +2753,13 @@ static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
     dst->flags      = src->flags;
     dst->sequence   = src->sequence;
 
+    if (src->hwaccel_picture_private) {
+        dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
+        if (!dst->hwaccel_priv_buf)
+            goto fail;
+        dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
+    }
+
     return 0;
 fail:
     ff_hevc_unref_frame(s, dst, ~0);
@@ -2865,12 +2769,10 @@ fail:
 static av_cold int hevc_decode_free(AVCodecContext *avctx)
 {
     HEVCContext       *s = avctx->priv_data;
-    HEVCLocalContext *lc = &s->HEVClc;
     int i;
 
     pic_arrays_free(s);
 
-    av_freep(&lc->edge_emu_buffer);
     av_freep(&s->md5_ctx);
 
     av_frame_free(&s->tmp_frame);
@@ -2881,17 +2783,17 @@ static av_cold int hevc_decode_free(AVCodecContext *avctx)
         av_frame_free(&s->DPB[i].frame);
     }
 
-    for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
-        av_freep(&s->vps_list[i]);
-    for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
-        av_buffer_unref(&s->sps_list[i]);
-    for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
-        av_buffer_unref(&s->pps_list[i]);
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.vps_list); i++)
+        av_buffer_unref(&s->ps.vps_list[i]);
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++)
+        av_buffer_unref(&s->ps.sps_list[i]);
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.pps_list); i++)
+        av_buffer_unref(&s->ps.pps_list[i]);
 
-    for (i = 0; i < s->nals_allocated; i++)
-        av_freep(&s->nals[i].rbsp_buffer);
-    av_freep(&s->nals);
-    s->nals_allocated = 0;
+    for (i = 0; i < s->pkt.nals_allocated; i++)
+        av_freep(&s->pkt.nals[i].rbsp_buffer);
+    av_freep(&s->pkt.nals);
+    s->pkt.nals_allocated = 0;
 
     return 0;
 }
@@ -2924,7 +2826,7 @@ static av_cold int hevc_init_context(AVCodecContext *avctx)
     if (!s->md5_ctx)
         goto fail;
 
-    ff_dsputil_init(&s->dsp, avctx);
+    ff_bswapdsp_init(&s->bdsp);
 
     s->context_initialized = 1;
 
@@ -2957,26 +2859,35 @@ static int hevc_update_thread_context(AVCodecContext *dst,
         }
     }
 
-    for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
-        av_buffer_unref(&s->sps_list[i]);
-        if (s0->sps_list[i]) {
-            s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
-            if (!s->sps_list[i])
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.vps_list); i++) {
+        av_buffer_unref(&s->ps.vps_list[i]);
+        if (s0->ps.vps_list[i]) {
+            s->ps.vps_list[i] = av_buffer_ref(s0->ps.vps_list[i]);
+            if (!s->ps.vps_list[i])
+                return AVERROR(ENOMEM);
+        }
+    }
+
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++) {
+        av_buffer_unref(&s->ps.sps_list[i]);
+        if (s0->ps.sps_list[i]) {
+            s->ps.sps_list[i] = av_buffer_ref(s0->ps.sps_list[i]);
+            if (!s->ps.sps_list[i])
                 return AVERROR(ENOMEM);
         }
     }
 
-    for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
-        av_buffer_unref(&s->pps_list[i]);
-        if (s0->pps_list[i]) {
-            s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
-            if (!s->pps_list[i])
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.pps_list); i++) {
+        av_buffer_unref(&s->ps.pps_list[i]);
+        if (s0->ps.pps_list[i]) {
+            s->ps.pps_list[i] = av_buffer_ref(s0->ps.pps_list[i]);
+            if (!s->ps.pps_list[i])
                 return AVERROR(ENOMEM);
         }
     }
 
-    if (s->sps != s0->sps)
-        ret = set_sps(s, s0->sps);
+    if (s->ps.sps != s0->ps.sps)
+        ret = set_sps(s, s0->ps.sps);
 
     s->seq_decode = s0->seq_decode;
     s->seq_output = s0->seq_output;
@@ -2998,7 +2909,7 @@ static int hevc_decode_extradata(HEVCContext *s)
 {
     AVCodecContext *avctx = s->avctx;
     GetByteContext gb;
-    int ret;
+    int ret, i;
 
     bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
 
@@ -3055,6 +2966,16 @@ static int hevc_decode_extradata(HEVCContext *s)
         if (ret < 0)
             return ret;
     }
+
+    /* export stream parameters from the first SPS */
+    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++) {
+        if (s->ps.sps_list[i]) {
+            const HEVCSPS *sps = (const HEVCSPS*)s->ps.sps_list[i]->data;
+            export_stream_params(s->avctx, &s->ps, sps);
+            break;
+        }
+    }
+
     return 0;
 }
 
@@ -3105,6 +3026,14 @@ static void hevc_decode_flush(AVCodecContext *avctx)
 
 #define OFFSET(x) offsetof(HEVCContext, x)
 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
+
+static const AVProfile profiles[] = {
+    { FF_PROFILE_HEVC_MAIN,                 "Main"                },
+    { FF_PROFILE_HEVC_MAIN_10,              "Main 10"             },
+    { FF_PROFILE_HEVC_MAIN_STILL_PICTURE,   "Main Still Picture"  },
+    { FF_PROFILE_UNKNOWN },
+};
+
 static const AVOption options[] = {
     { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
@@ -3131,6 +3060,7 @@ AVCodec ff_hevc_decoder = {
     .flush                 = hevc_decode_flush,
     .update_thread_context = hevc_update_thread_context,
     .init_thread_copy      = hevc_init_thread_copy,
-    .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
-                             CODEC_CAP_FRAME_THREADS,
+    .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
+                             AV_CODEC_CAP_FRAME_THREADS,
+    .profiles              = NULL_IF_CONFIG_SMALL(profiles),
 };